Energy Policy 188 (2024) 114092

Contents lists available at ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

The Iberian exception: Estimating the impact of a cap on gas prices for
electricity generation on consumer prices and market dynamics
Manuel Hidalgo-Pérez a ,∗, Natalia Collado b , Jorge Galindo b , Ramón Mateo c
a
Universidad Pablo de Olavide, Ctra Utrera s/n, 41013, Sevilla, Spain
b
Universitat Ramon Llull, Esade, Spain
c
BeBarlet, Spain

ARTICLE INFO ABSTRACT

JEL classification: As the volatility of short-term energy market prices increases due to exogenous shocks and the changing
D12 nature of the energy mix, market interventions are gaining importance in the policy debate. Accurate and
Q41 robust quantification of their impact is becoming therefore essential. This paper conducts a causality analysis
Q48
to evaluate the Spanish ‘‘gas cap’’ for electricity generation during the 2022 energy crisis. We use Bayesian
Keywords: structural time series models to isolate its impact on affected consumers, primarily those under the regulated
Policy evaluation tariff, from June to December 2022. Our results show that the mechanism successfully lowered prices compared
Gas price cap to a counterfactual scenario without the intervention. However, the policy also led to an increase in electricity
Consumer prices
generation from gas-fired combined cycle plants. In addition, exports to France increased by over 80% after
Gas consumption
implementation, as Spanish wholesale prices became cheaper than French prices. Our analysis (1) provides
an accurate quantification to date of one of the most consequential energy market interventions in recent
years; (2) demonstrates the fruitful use of a novel evaluation method for energy policy; (3) highlights the
trade-off between the short-term goal of consumer price relief and the long-term goals of fossil fuel reduction
and decarbonisation.

1. Introduction
The energy landscape in Europe has experienced a turbulent time
in recent years, marked by the coincidence of several crises. Begin-
ning with the outbreak of the global pandemic and followed by the
geopolitical shockwaves of the Russian invasion of Ukraine and the
subsequent aim of reducing European energy dependence from Russia,
the continent has faced an unprecedented surge in energy prices.
These were especially intense for natural gas, given the continent’s
dependence on the source from foreign partners, especially Russia itself,
which accounted for almost half the imports of this fossil fuel. As a
result, in December 2021, the price of the Dutch TTF1 was ten times
higher than in 2019.
This escalation shook the European electricity markets, especially
the day-ahead markets, due to two interacting factors: (1) the use
of natural gas for electricity generation and (2) its usual role as a
marginal technology in this market. Combined cycle power plants pro-
vide flexibility in the system by meeting peak demand when renewable
generation is low. During these hours, they are the most expensive
technology and determine the market price. As gas is a necessary input
for their production, any increase in its price is immediately reflected in
their bids and consequently in the price of electricity. Moreover, since
it takes on average two MWh of gas to produce one MWh of electricity,2
the electricity market amplifies the impact of a price shock for this fossil
fuel, as shown in Fig. 1.
Faced with increasing economic and social hardship caused by
escalating electricity prices, national governments and the European
Commission proposed emergency measures to reduce these prices and
mitigate their negative effects. Several alternatives were discussed that
had a direct impact on the functioning of the day-ahead market, such as
setting a price cap, limiting the selling price of low-margin technologies
or compensating fossil-fuelled power generators (European Commis-
sion, 2022). However, none of these measures were ultimately adopted
at the EU level. Instead, the supranational focus shifted to initiating

∗ Corresponding author.
E-mail addresses: mhidper@upo.es (M. Hidalgo-Pérez), natalia.collado@esade.edu (N. Collado), jorge.galindo@esade.edu (J. Galindo),
ramon.mateo@bebartlet.com (R. Mateo).
1
The Dutch TTF, or Title Transfer Facility, is a system registering the delivery of gas in the Dutch gas system and also the main index used in Europe for
long-term contracts.
2
This ratio may vary depending on the efficiency of each power plant.

https://doi.org/10.1016/j.enpol.2024.114092
Received 2 November 2023; Received in revised form 4 March 2024; Accepted 18 March 2024
Available online 28 March 2024
0301-4215/© 2024 Published by Elsevier Ltd.
M. Hidalgo-Pérez et al.
Fig. 1. Evolution of wholesale electricity and gas prices in Europe (in e/MWh).
Source: ESIOS, Investing.com.
electricity market reform and implementing fiscal measures, includ-
ing reducing VAT or excise taxes, regulating retail prices, providing
transfers to vulnerable consumers and supporting industry and taxing
windfall profits (Sgaravatti et al., 2021).
Despite those, the impact of soaring electricity prices in Spain
gained particular attention from policymakers and the general public.
According to the Spanish National Markets and Competition Commis-
sion (CNMC for its Spanish acronym),3 in early 2022 around 10 million
households were under the regulated tariff, so-called Voluntary Price
for the Small Consumer (PVPC for its acronym in Spanish, VPSC for its
English acronym henceforth). Despite its regulated nature, the VPSC is
linked to hourly wholesale market results through a dynamic pricing
model. For consumers under the VPSC scheme, this means that their
monthly electricity bill is a direct reflection of market conditions.
However, for the 18 million households on fixed tariffs in the free
segment of the retail market, prices are only updated once a year when
their contracts with their suppliers come up for renewal.
This heightened concern was compounded by two additional fac-
tors: (1) the VPSC is the tariff for many vulnerable households, as it is a
prerequisite for access to fuel poverty assistance,4 and (2) its significant
impact on the national Consumer Price Index (CPI), which by design
was more exposed to increases in wholesale electricity prices.5
The combination of these factors led to the introduction, from June
2022, of a mechanism to reduce the day-ahead wholesale electricity
3
See CNMC (2022).
4
This support consists of a discount on the regulated tariff.
5
Traditionally, the Spanish National Statistics Institute has only taken the
regulated tariff into account when measuring electricity prices. However, after
noting that there was overexposure during the energy crisis, since January
2023 it has also included information on contracts that do not fall under the
regulated tariffs.
2
Energy Policy 188 (2024) 114092
prices, and therefore the final bill for consumers with VPSC, by de-
coupling them from natural gas prices through a partial modification
of the market’s marginal system. This paper focuses on assessing the
impact of such a mechanism, referred to as the ‘‘Iberian exception’’
or ‘‘gas cap’’, although this is technically incorrect since it is not an
actual cap, as explained in Section 2. Even though it was introduced by
the Spanish and Portuguese governments and thus applies to the entire
Iberian electricity market (MIBEL), the analysis focuses on its impact in
Spain. To assess whether the policy achieved its stated goal, we conduct
a causality analysis to isolate the effect of the cap from other factors. In
Section 3 we explain in detail our method, departing from a model that
can approximate the time series of the wholesale electricity price in the
regulated market, and using it to feed Bayesian structural models able
to produce a counterfactual of the wholesale electricity price without
implementing the cap. In Section 4 we compare the counterfactual
with the observed time series to measure the impact of the regulatory
measure.
Approving and implementing the mechanism required convincing
other European partners and the European Commission itself, who
voiced two tangible worries aside from the more general alert of the
exception risking ‘‘a distorsion on the single internal market’’ (Inde-
pendiente, 2022): (1) whether it could incentive gas consumption,
thus undermining energy savings and decarbonisation goals, leading
to an actual warning by the Commission in October 2022 (Arroqui,
2022); and (2) whether the lowering of the prices could ‘‘leak’’ to
other European countries, e.g. France, exposing the Spaniards to a
potential subsidy for French consumers. Sections 5 and 6 look for proof
of these outcomes extending our causal impact methodology. Section 7
concludes by highlighting what we believe is our core contribution plus
indicating where further research could lead us.
M. Hidalgo-Pérez et al.
2. The measure and existing evidence on it
The day-ahead wholesale electricity market has a marginalist de-
sign, i.e. it operates on the principle that the price of electricity is set
by the most expensive energy source needed to meet current demand.
Typically, this marginal source is a combined cycle power plant, which
has higher operating costs due to its required inputs (e.g. natural gas).
On the other hand, cleaner technologies such as solar, wind, hydro and
nuclear power are the first ones to be dispatched due to their lower
marginal costs. When demand exceeds what the baseline supply of
these sources can provide and gas-fired power plants enter the market,
any increase in the price of this fossil fuel is reflected in the wholesale
price of electricity.
However, the price dynamics in the wholesale electricity market do
not affect all Spanish consumers in the same way, due to the different
tariffs in the retail market. Consumers on unregulated fixed tariffs
experience price changes mainly during the annual renegotiation and
renewal of their contracts. Conversely, households on the regulated
tariff (VPSC), which is indexed to spot prices, are directly affected
by short-term fluctuations in the wholesale market. This direct link
exposes them to immediate price volatility, in contrast to the delayed
impact felt by consumers on fixed tariffs. What is more, there are often
many vulnerable consumers on the regulated tariff, as this is a condition
for access to the Electricity Social Bonus - a discount on their bills
designed to help combat energy poverty.
To address this short-term risk while minimising the impact on
public finances and the design of electricity bills, in mid-June 2022,
Spain and Portugal introduced a mechanism informally and widely
known as the ‘‘gas cap’’. In parallel, Spain also started a process to
reform the VPSC to reduce its volatility by indexing the tariff not only
to the spot market but also to the futures markets. The new regulated
tariff structure officially entered into force in January 2024.
The ‘‘gas cap’’ mechanism aimed to reduce the impact of wholesale
natural gas prices to wholesale electricity market pricing. It consisted
of compensation to specific electricity generation facilities using fossil
energy sources in exchange for limiting their bids in the wholesale
electricity market, thereby reducing the market price. Eligible plants
included natural gas combined cycle power plants (CCGT), coal-fired
thermal plants and certain combined heat and power (CHP) plants. The
compensation was calculated as the difference between two gas prices:
the Iberian wholesale market price (MIBGAS), and a reference price
set by the new standard. The reference price was set at e40/MWh for
the first six months and increased by e5/MWh each month thereafter
until it reached e70/MWh at the end of the initial one-year application
period.6 The resulting compensation was disseminated by the market
operator since its enactment, and the affected power plants included
it in their energy price offers on the wholesale market. It therefore
reduced the price by the difference between the market price and the
pre-set price. This mechanism of internalising the compensation acted
as a de facto cap on the cost of gas, which was passed on in the final
price of electricity.
Due to the marginalist functioning described above, the mecha-
nism acted as a limit on the benefits that inframarginal technologies
(i.e. those that are expected to enter the market at a lower bid price,
such as solar, wind, hydro and nuclear) would end up receiving. While
the measure was in place, and until 31 December 2023, Spain also had
a limit on the ‘‘windfall profits’’ that these technologies could receive.7
6
The Iberian exception was later on extended until December 2023 with
new reference prices. See the Royal Decree outlining the extension of the
mechanism: Real Decreto-ley 3/2023, de 28 de marzo, de prórroga del
mecanismo de ajuste de costes de producción para la reducción del precio
de la electricidad en el mercado mayorista regulado en el Real Decreto-ley
10/2022, de 13 de mayo.
7
See Royal Decree-Law 17/2021.
3
Energy Policy 188 (2024) 114092
The cost of financing the compensation paid through this mecha-
nism is undermined by the congestion rents generated by the inter-
connection with France, since lower wholesale prices would increase
both the difference between Spanish and French prices and export flows
after the mechanism is applied.8 After deducting these from the costs,
the remainder is distributed as a payment obligation on the wholesale
market demand. At market close, the adjustment is paid to eligible
generators to ensure that their marginal costs are fully recovered and
the costs incurred are passed on to consumers as part of their electricity
bills. This pass-on initially affected consumers at the regulated end
of the retail market, but once their contracts have been updated,
consumers on the free market will also internalise the costs.
However, even though consumers bear the cost of the compensation
in their bills, the way the mechanism is designed, they should make net
savings. This is because the remuneration foregone by inframarginal
technologies is less than the compensation paid. The difference leads to
lower clearing prices than if the mechanism were not in place, resulting
in higher consumer surplus. Brito (2022) illustrates this comparison in
Fig. 2. In the right panel, the situation without the cap is depicted,
where the yellow area represents consumers’ surplus, defined as the
difference between the market price (100e/MWh) and their maximum
willingness to pay for each unit consumed (the blue line). With the cap
in place, the market price is supposed to decrease, allowing consumers
to increase their surplus even with the compensation payment (red area
in the left panel).
The critical question is therefore empirical: does this hold up in
reality, once we look at actual prices under the new policy regime? San-
cha (2022), examining the impact of the mechanism during the first
200 days of its application, concludes that the average reduction in
the wholesale marginal price during the analysis period was significant
at 123e/MWh. To estimate the counterfactual scenario, the study
assumed inelastic demand and calculated the counterfactual electricity
price as the actual price plus the subsidy that power plants receive. The
author also analysed the impact on generators and concluded that, al-
though the reduction in the wholesale price reduced the extraordinary
profits of the inframarginal power plants that did not enter into forward
contracts, the fossil fuel power plants did not suffer any loss of revenue,
as this was offset by the contributions from MIBEL demand and conges-
tion revenues from the interconnection between Spain and France. The
study also showed that consumers’ experiences varied depending on the
type of contract. For example, consumers with regulated and market-
indexed contracts benefited from a 17% reduction in the energy price
component of their bill due to the intervention, while consumers with
fixed price contracts experienced different scenarios depending on the
date and structure of their contract renewal.
Other researchers assessed the impact of the intervention based on
an econometrically modelled counterfactual scenario, i.e. an estimate
of how the variable in question would have evolved in the absence
of the policy. In Salas et al. (2022), the focus is on the impact of
the intervention on the VPSC. They estimate a counterfactual that
models this price with auto-regressive errors and controls for weekly
seasonality. The explanatory factors of the model included the gas
price, climatic variables such as wind and solar conditions, and the
wholesale electricity price in France. The results of this analysis showed
a cumulative average reduction in VPSC of 20.7% by the end of Septem-
ber 2022. Robinson et al. (2023) construct counterfactual supply and
demand curves for the Iberian electricity market to assess the impact of
the mechanism on Spanish consumers. To this end, the authors estimate
a counterfactual supply curve representing the offers that electricity
generators would have made in the absence of the intervention based
8
See the Royal Decree that outlines the functioning of the mechanism: Real
Decreto-ley 10/2022, de 13 de mayo, por el que se establece con carácter
temporal un mecanismo de ajuste de costes de producción para la reducción
del precio de la electricidad en el mercado mayorista.
M. Hidalgo-Pérez et al.
Fig. 2. Representation of potential savings for the consumer.
Source: La Excepción ibérica a debate ¿Una oportunidad
perdida? Brito, P (2022).
on the actual offers made by generators during the first 100 days of im-
plementation, together with the actual compensation received by fossil
fuel generators. Real data reflecting Iberian demand offers were also
used for the demand side, and actual demand contributions to finance
the compensation were added to incorporate demand elasticity into
the modelling. They find that under the assumption that only Iberian
large consumers would have responded to the price change, the savings
would have been 13%. In addition, the study considers the impact of
French demand on Iberian market prices and interconnections. Under
this alternative counterfactual assumption, the authors suggest that
affected consumers, i.e. those subject to the VPSC or whose retail prices
are linked to the wholesale price, would have paid less if the Iberian
exception had not been introduced.
Our work places itself within these empirical efforts and it does so
by adding two distinct elements. First and foremost, we believe that
our method complements and extends the aforementioned analyses:
using a causal impact methodology based on a Bayesian Structural
Time Series model allows us not only to estimate a counterfactual
estimate but also to perform posterior inference and assign a proba-
bility that the observed effect is a direct result of the intervention. In
fact, to the best of our knowledge, ours is one of the first examples
of a fruitful use of this novel evaluation method for a price-based
energy policy intervention. Furthermore, we do not only measure the
effects on prices but also on gas consumption for electricity generation
and exports of electricity to France. By quantifying this dimension
along with consumption savings, our contribution aims to go beyond
assessing the unintended effects of a particular policy: we show how
policies that affect price signals activate the trade-off between the
short-term goal of reducing consumer prices and the long-term goals of
decarbonisation.
3. Our method to measure the cap’s actual impact
As stated above, our primary goal is to determine whether the
VPSC experienced a price reduction since its onset. Additionally, it
4
Energy Policy 188 (2024) 114092
is imperative to investigate whether this price reduction, if indeed it
occurred, persisted throughout the intervention. Operationalised, we
aim to measure whether the intervention had a causal effect on our
target time series, which is the VPSC price.
In recent years, significant progress has been made in the field
of causal analysis, particularly in economics. Methodologies such as
differences-in-differences, synthetic indicators and, more recently, ad-
vances based on machine learning techniques, have transformed public
policy evaluation.9 While many of these techniques were originally
designed for cross-sectional or panel data to analyse causal relation-
ships between a treatment and a variable of interest within a specific
population, they have been adapted for time series analysis. The causal
impact method, introduced by Google Inc. and described by Brodersen
et al. (2015) (BGKNS, hereafter), is a recently developed methodology
for examining how variables evolve in response to treatment using
Bayesian Structural Time Series (BSTS) models.
Technically, the BSTS method extends the differences-in-differences
approach to time series analysis by explicitly defining a structural
model to predict the counterfactual of a time series both before and
after an intervention. This modelling framework enables the estimation
of the treatment effect on one or more series included in the analysis.
Essentially, the effect is quantified by comparing the expected perfor-
mance of the time series with its observed behaviour, thus capturing
the treatment effect on the treated. To estimate the counterfactual time
series, the method combines information from predictor variables in
what is commonly referred to as a synthetic control (Abadie et al.,
2010).
The construction of this synthetic control involves considering var-
ious sources of information, including the performance of the target
time series before the treatment and its response to the intervention.
Furthermore, it incorporates data from other time series that may
9
See Hernán and Robins (2023) for a survey of this progress and
applications.
M. Hidalgo-Pérez et al.
predict the behaviour of the target series after the treatment, while
ensuring that these variables are not influenced by the treatment itself.
Additionally, it uses a Bayesian framework underlying so that the model
estimated utilises prior knowledge about the model parameters and
subsequently performs posterior inference on the counterfactual.
Implementing the BSTS involves several steps. Firstly, we define
the structural model that approximates the historical evolution of the
relevant variables for the counterfactual analysis. This process entails
analysing the economic nature of the target variable (𝑦𝑡 ), VPSC in
our case, and identifying suitable covariates to be included in the
model. Crucially, the selected covariates and the model must meet
specific criteria, such as exogeneity for the control variables used in
constructing the synthetic control.
Specifically, the BTST model belongs to the class of state-space mod-
els for time series and is represented by the following two equations:
𝑦𝑡 = 𝑍𝑡𝑇 𝛼𝑡 + 𝛽𝑋𝑡 + 𝜀𝑡 (1)
𝛼𝑡+1 = 𝑇𝑡 𝛼𝑡 + 𝑅𝑡 𝜂𝑡 (2)
where
𝜀𝑡 ∼ 𝑁(0, 𝜎𝑡2 )
(3)
𝜂𝑡 ∼ 𝑁(0, 𝑄𝑡 ).
The Eq. (1) is the observational equation within a state space model.
This equation links the data we observe, a series of one variable 𝑦𝑡
with a set of k-dimensional latent (states) variables 𝛼𝑡 and covariates
defined by 𝑋𝑡 . The Eq. (2) is the state equation or transition equation,
which explicit the evolution of these latent variables through time. 𝑍𝑡𝑇
is a vector of dimension 𝑘𝑥1 called output matrix, 𝑇𝑡 is a matrix of
dimension 𝑘𝑥𝑘 called transmission matrix, while 𝑅𝑡 is a control matrix
of 𝑘𝑥𝑝 dimension. Finally, 𝑅𝑡 𝜂𝑡 implies, as is explained by BGKNS, the
possibility of incorporating state components of less than the full rank.
All of these matrices contain unknown parameters and known values
which are often set as 0 and 1.
By varying the matrices 𝑍, 𝑇 , 𝐺 and 𝑅 and defining different latent
variables we can model several distinct behaviours for the time series
(including the more well-known such as ARMA or ARIMA). In this
exercise, we interpret the model previously presented by the following:
𝑦𝑡 = 𝜇𝑡 + 𝛾𝑡 + 𝛽𝑋𝑡 + 𝜖𝑡 (4)
𝜇𝑡+1 = 𝜇𝑡 + 𝜂𝜇,𝑡 (5)
∑
𝑆−1
𝛾𝑡+1 = − 𝛾𝑡 + 𝜂𝛾,𝑡 (6)
𝑠=1
In this particular scenario, 𝑍𝑡𝑇 , 𝑅𝑡 , and 𝑇𝑡 are equal to one, while
𝛼𝑡 is reduced to two latent variables (𝑘 = 2), a random variable
denoted by 𝜇𝑡 , governed by a random walk model and is conventionally
referred to as the ’local level’ component and increases independently
as the other parameters do not contribute significantly to explaining
the data, and 𝛾𝑡 variables, which model seasonal components, and in
our model we include two of them specifically representing weekdays
and months (𝑆 = 7 and 𝑆 ′ = 12).10 This choice is motivated by
structural components associated with different days of the week and
the traditional seasonality in a year in electricity price data.
The linear regression is presented using covariates 𝑋𝑡 that further
help to explain observed data. The better this component works in the
prediction task, the lower the local level component should be. Our
objective, therefore, is to minimise the role of the latent variable in such
10
We must think of 𝛾𝑡 as a regression with seasonal dummy variables, in
this case, with S = 7 with six dummy variables to capture the week seasonal
cycle.
5
Energy Policy 188 (2024) 114092
a way that the covariates in 𝑋𝑡 explain the largest possible proportion
of the evolution of the variable 𝑦𝑡 .
Finally, the parameters 𝜖𝑡 and 𝜂𝜇,𝑡 represent noises which are related
to measuring 𝑦𝑡 and 𝜇𝑡 and they follow a normal distribution with zero
means and 𝜎𝜖 and 𝜎𝜂 standard deviation respectively.
To estimate the model, the first step, following the Bayesian ap-
proach, involves specifying prior distributions for each model param-
eter. In this case, the initial step is to define the prior distributions
for the errors, which entails assigning values to the error variances.
The second step leads to the acquisition of posterior distributions.
To obtain them, and given the Eqs. (4) to (6) numerical methods
based on simulation techniques must be employed. Specifically, for the
estimation of our model, we will use a Monte Carlo Markov Chain
algorithm (Hamiltonian Monte Carlo). Once the model is estimated, we
obtain a posterior distribution of the counterfactual time series. Finally,
we subtract the predicted from the observed response during the post-
intervention period, giving us a semiparametric Bayesian posterior
distribution for the causal effect.
4. Estimating the impact of the gas cap on final electricity prices
To assess the measure, we must focus on the main stated objective
of the gas cap: reduce the final bill of the regulated market consumer.
Then, the logical course of action is to examine whether, since its
enactment, there has indeed been a significant reduction in the VPSC.
However, the cap may have other effects, which have been observed
since it came into operation and are also worth evaluating, such as a
possible increased use of combined cycle power plants or an increase in
exports to France. Both outcomes would result in a rise in gas consump-
tion. Consequently, we believe it is necessary to expand the analysis to
these two areas to enrich the present and future debate regarding the
measure or potential reforms. In this section we confine ourselves to
the first objective and leave the rest to the following sections.
Fig. 3 depicts the evolution of the daily VPSC series11 in Spain
from January 1, 2020 to December 31, 2022. With the onset of the
energy crisis, prices increased markedly from the summer of 2021,
diverging from the historical pattern. As explained in Hidalgo-Pérez
et al. (2022), the marginalist design of the wholesale electricity market
and the indexation of the VPSC to it, as described in the introduction,
is the main reason for this increase, as electricity generated from gas
and other fossil fuels typically sets the daily price.
The cap should lead to a lower VPSC, partially altering the marginal-
ist market system, from June 15, 2022 onward. To estimate the effect of
the gas cap, we require a model that best approximates the historical
evolution of prices (defined by Eq. (1)). This process comprises two
steps:
1. First, we select the variables (covariates) that, together with a
model like (1), enable us to extrapolate the series shown in Fig. 3
with minimal error.
2. Next, we estimate a BSTS model to obtain the counterfactual
scenario and a posterior probability that the enactment of the gas
cap could reduce the VSPC, obtaining a measure of ascertaining
the likelihood that it is or is not due to the policy.
4.1. Step 1: The hypothetical price series
Thus, the first step should focus on designing a model, together
with selecting appropriate variables, that enables predicting the VPSC
as closely as possible over the available time series. To accomplish
this, we begin by determining which gas price series to utilise as a
11
In addition to energy costs, the VPSC price series also includes tolls and
charges, which are regulated prices designed to cover the costs of transmission
and distribution networks and other regulated costs.
M. Hidalgo-Pérez et al. Energy Policy 188 (2024) 114092

Fig. 3. Evolution of VPSC.

Source: Red Eléctrica.

Fig. 4. Evolution of gas price: MIBGAS y TTF.

Source: Own elaboration based on data from
Red Eléctrica and REFINITIV.

benchmark for estimating the VPSC. This choice is critical given the
current existence of two possible references: the gas price on the virtual
market in the Netherlands (Dutch TTF) and the price on the Iberian Gas
Market (MIBGAS). Objectively, there is no evidence for a final decision,
as the gas supply of combined cycle power plants could be linked to one
of the two prices and this is private information. Therefore, we initially
let the data dictate: our primary selection criterion was that the gas
price used to build the model should be the one that best approximated
the VPSC price series statistically.
For most of the period where both gas prices are available, it is
observed that the TTF and MIBGAS have evolved similarly (Fig. 4).
However, this parallel evolution diverges from 4 April 2022. One
potential explanation is the enactment of sanctions against Russia a
few days later, which both markets had already anticipated by that
date. Thus, the increase in uncertainty over gas supply to Europe was
less pronounced for the Iberian Peninsula due to the existence of more
diversified channels for gas supply outside the Russian orbit. It is in
the divergence described over this period that we find the opportunity
to carry out a quantitative exercise that helps to discriminate between
the two prices in terms of their ability to emulate the evolution of the
VPSC.
To determine which gas price more closely tracks the evolution of
the VPSC, we conducted a quantitative modelling exercise. We split the
data into a training set from September 9, 2021, to April 4, 2022, and
a test set from April 5, 2022 to June 14, 2022. We then built two linear
regression models to predict the VPSC price — one using the Dutch TTF
gas price and another with the MIBGAS price. The models were trained
and then used to generate predictions on the test data. We evaluated
the predictions using root mean squared error (RMSE). Once the two
models were estimated we obtained that the TTF model achieved a
RMSE of 39.15 while the MIBGAS model had a RMSE of 53.64. Based
on the lower RMSE, we can conclude that the TTF gas price enables
more accurate predictions of the VPSC over the study period.
Therefore, TTF should be the price featured as a covariate in the
structural model. However, although the data indeed discriminates in
favour of TTF, during the year before the intervention MIBGAS has
growingly increased its share of new contracts within the Spanish gas
and electricity system. It may then be appropriate to test alternative
price series resulting from a weighting of TTF and MIBGAS prices, as
the two series follow different paths. To evaluate this, we test gas price
series that result from applying a continuum of weightings ranging from
0% to MIBGAS (therefore 100% to TTF) to 100% to MIBGAS (0% to
TTF). For each option, the mean squared error has been calculated and
is represented in Fig. 5.
Again, it appears that the highest explanatory power of the price
series is achieved with the TTF price series. The evolution of RMSE is

6
M. Hidalgo-Pérez et al.
Fig. 5. Mean squared error of VPSC prediction by different weights to gas prices.
Source: Own elaboration.
monotonic concerning the weight assigned to TTF, with the minimum
value occurring when this weight is set to 100 for this gas price series.
However, the growing presence of MIBGAS on contracts during
2022 should be reflected in our analysis. As a balanced solution, given
that this gain in the share of MIBGAS in 2022 reached 20%, we believe
that using this weight to obtain a weighted price series from April 2022
is appropriate at a low potential loss based on what is represented in
Fig. 5, and in compensation further shielding our results from unob-
servable inaccuracies. We thus follow a 80/20 weight for TTF/MIBGAS
starting when both prices are available, i.e., from September 7, 2021.
As long as both series have evolved similarly, there will be no difference
between this weighted price and either of the two separately. By the
time the two series separate, the weighted series will be nonetheless
substantially closer to TTF, ensuring that TTF takes precedence over
MIBGAS, but still allowing the aforementioned extra safe room.
4.2. Step 2: estimating the effect of the cap on the VPSC
In the next step, we must choose which other variables will accom-
pany the gas prices for the model estimation. Several possibilities arise
here that must be taken into account for a proper estimation of the
causal effect.
First, we estimate three groups of models, each including one of the
gas prices (TTF, weighted average of TTF, and MIBGAS prices with lags
of 1 and 2 days). Second, an autoregressive variable of order one (the
VPSC price itself with a lag of 1 day) aims to capture the structure of
the time series. Third, two interventions corresponding to the reforms
of June 1, 2021 (reform 1) and September 15, 2021 (reform 2) could
have a significant impact on the price. The first relates to a change in
the so-called ‘‘tolls’’, regulated prices designed to cover the costs of the
transmission and distribution networks and set by the Spanish national
regulator. On 1 June 2021, the different tolls were unified so that all
domestic consumers pay the same price, which will also include hourly
discrimination for power and energy. The second reflects a reduction
in fees, regulated prices set by the Ministry of Ecological Transition
and the Demographic Challenge to cover other regulated costs of the
system, which began on 15 September 2021. Additionally, we include
controls for days of the week and months of the year to account for
seasonality.
From this point, we introduce two different sets of variables that
control for the various technologies used to produce electricity, pro-
viding two distinct estimations of the causal effect. In the first group,
we use the daily production volumes of wind and solar power plants
to consider the influence of the supply mix in price determination.
The reason why we use the production volumes of wind and solar
7
Energy Policy 188 (2024) 114092
power plants as explanatory variables and not other technologies such
as combined cycle or CHP plants is simple. We assume that the for-
mer, especially wind and solar power plants, are exogenous to market
conditions, as their market entry depends mainly on meteorological
factors. As explained in Section 3, the variables included must be
independent of the measure or intervention whose impact is being
assessed for the BSTS model to work properly. Using the production
data of these plants instead of gas plants implies that the reasons for
their operation go beyond market conditions and are more related to
the availability of their environmental factors. However, as a robustness
check, we estimate a second set of models, replacing the production
of wind and solar plants with daily average meteorological variables
for Spain. Specifically, based on AEMET data, we have included in our
analysis, following Mosquera-López et al. (2017) data on average tem-
perature, precipitation, minimum and maximum temperature, average
wind speed, maximum wind gust, as well as insolation. Furthermore,
by eliminating production by technologies, we can include, to control
for market dynamics, the total energy demand for each day, captured
by the total production of electrical energy. This provides us with six
different models to estimate the causal effect of the gas cap on the
evolution of electricity prices.
The results of the OLS regressions between VPSC and the different
gas prices (with a combination of the covariates mentioned above) are
shown in Tables A.1 and A.2 located in the appendix. The regressions
suggest that VPSC could be well approximated with only a few re-
gressors and that the best model is obtained when the TTF price is
used alone or in combination with MIBGAS (with the above weights).
Moreover, the ability of the models to approximate the daily evolution
of the VPSC since January 2014 is very high, above 94% if TTF the same
if weighting is used and 90% if MIBGAS is used. So, once it is proven
that it is possible to develop a model that can accurately approximate
(and predict) the evolution of the price series, the next step is to use
these models to analyse the impact of the gas cap intervention on
prices.
The estimated causal effect varies depending on the chosen model,
although it remains highly significant (Fig. 6). Additionally, in all six
models (three incorporating wind and solar photovoltaic plant produc-
tion data in blue and three using meteorological and total generation
data in red), the likelihood of observing the effect in the VPSC without
the gas cap is extremely low: the one-sided Bayesian tail-area proba-
bility for all models is 0.0%. Consequently, the causal effect can be
deemed statistically significant in all instances.
In view of this and depending on the reference gas price used, the
effect of the electricity price reduction varies. Using the TTF or the
weighted gas price as reference implies a larger downward adjustment
M. Hidalgo-Pérez et al.
Fig. 6. Estimated causal effect on VPSC for each model — 90% confidence interval.
Note: The blue results refer to the models that include electricity generation from wind and solar plants, while the red ones pertain to the use of meteorological variables and the
total system generation. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Source: Own elaboration.
of electricity prices, between 20% and 28%, consistent with those
of Salas et al. (2022) who estimate a cumulative saving of 20.8%, while
using MIBGAS as reference implies a smaller effect, between 7% and
18%. These differences result from taking into account a price such as
TTF, which was higher than MIBGAS, which was significantly lower,
during the period of the gas cap studied here. If the Spanish combined
cycle power plants had the Iberian gas price as a reference, the savings
would have been much lower.
Fig. 7 shows the estimated savings in euros as the difference be-
tween the lines shown in Fig. 3 and the counterfactual estimated by
the different BSTS models using wind and solar photovoltaic plant
production data.12 Until 14 June, the ‘‘savings’’ fluctuate around zero,
indicating that we are dealing with a random error corresponding to a
model that meets the minimum requirements for its correct estimation.
However, from 15 June onwards, all lines for each estimated model
diverge significantly into negative territory, suggesting that the gas cap
has led to an apparent reduction in the VPSC prices that customers pay
in this tariff. Moreover, this saving grew over the weeks, reaching a
maximum of almost 250 euros in the last week of August.
In monetary terms, the average saving since 15 June would be
around 130e/MWh. If we assume, based on the information collected
by the Spanish National Competition Authority’s household panel,
that an average household consumes about 8 kWh per day, we could
estimate the average household savings since 15 June to be about
206e. Assuming that around 9 million households were under the
VSPC (as estimated by the aforementioned Authority in May 202213 ),
the savings since its implementation would range close to 1.9 billions.
12
The results using meteorological data and total generation are very
similar; therefore, for the sake of space economy, they are not displayed.
13
See CNMC (2023).
8
Energy Policy 188 (2024) 114092
Considering that household expenditure on electricity exceeded 18
billions in 202214 these total savings are of relevance.
5. Effect on the use of the combined cycle
The gas cap may not only have affected price developments, but
also the use of the different technologies used to generate electricity in
the Spanish electricity system, and thus the potential savings that could
have been achieved through the measure. Therefore, the above analysis
was repeated in parallel to see if the introduction of the cap created
incentives to divert production to or away from certain technologies.
Given that the marginal system was originally designed to incentivise
the use of clean energy and discourage the use of more polluting
energy, it is important to assess whether and to what extent a change
as severe as this cap undermines this objective. This is particularly
important in the context on which the policy was enacted: European
countries were striving to reduce gas consumption both in terms of
independence and the energy transition, with the EU setting ambitious
targets for reducing consumption by 2022, up to 15% in some countries
(7% for Spain).
Intuitively, the application of the gas cap could create an incentive
for the use of a combined cycle plant. Before the cap came into
force, producers were exposed to rising gas prices and therefore had
to be more conservative in their production decisions. With the cap,
consumers are exposed to the risk of gas price increases through
compensation payments. Thus, combined cycle plants have a greater
incentive to bid in the market.
To quantitatively evaluate this possibility, we follow the same pro-
cedure used for the VPSC analysis. First, we define our underlying
14
Data extracted from the Spanish Household Budget Survey for 2022.
M. Hidalgo-Pérez et al.
Fig. 7. Estimated causal effect on VPSC for each model — 90% confidence interval.
Source: Own elaboration.
model, paying particular attention once again to our choice of ex-
planatory variables. The new dependent variable is the production in
combined cycle power plants, measured in gigawatt-hours (GWh). The
covariates remain the same as in the previous section, to which we
add another: hydropower generation. Its inclusion is motivated by its
unique market dynamics and potential influence on CCGT production.
Hydropower is typically supplied to the market based on its opportu-
nity cost, which becomes particularly relevant when more expensive
technologies, especially gas, are next in line for dispatch. Furthermore,
given the other covariates included in the model and the fact that
hydropower operational decisions are largely dependent on hydro-
logical conditions and not directly influenced by other unobservable
market factors affecting gas generation, hydropower generation can
be considered exogenous. In addition, we evaluate different models
using different prices, production in wind and solar photovoltaic farms,
meteorological variables and total generation. The results of the model
estimations are presented in Tables A.3 and A.4
Fig. 8 illustrates the impact on electricity generation in combined
cycle power plants as a result of the introduction of the gas cap.
Regardless of the price used in the analysis, we detect a significant and
positive increase using wind and photovoltaic production (blue ones).
However, the estimated size of the causal effect is slightly different,
ranging between 7% and 27%.
Fig. 9 shows the difference between the observed series of electricity
generation in GWh from combined cycle plants and what would have
been observed had the gas cap not existed. Its effects are apparent
already in the first months after its adoption. These results are con-
sistent with those of Eicke et al. (2022) in their critical assessment
of the export of the gas cap model to other EU countries: they note
that gas-fired electricity generation has increased by up to 42% in the
first few weeks, and highlight the tension between such increases and
the combined goal of reducing gas consumption and decarbonisation.
However, this effect seems to have disappeared within a few weeks of
the introduction of the Iberian gas cap and there does not seem to be
a significant impact on the use of combined cycle power plants in the
last months of 2022.
Nonetheless, it should be noted that the counterfactual estimate
assumes that electricity generation with this technology depends only
on exogenous factors, such as the market price, or the gas price, plus
photovoltaic and wind production. In reality, the greater or lesser use of
combined cycle power plants in a marginal wholesale electricity market
may also be affected by the use of other technologies beyond solar
and wind farms. In particular, it may depend on whether alternative
technologies are available and able to generate electricity on a given
day and at a given time. Although our analysis above shows a clear
impact of the cap on the use of combined cycle power plants, it is useful
to conduct a parallel analysis looking at these technologies.
Looking at what happened in Spain with alternative technologies,
a first thing to note is that, with the available information, it is
not possible to rule out that the lower use of some of these (such
as hydroelectric and, especially, cogeneration in favour of combined
cycle) is due to the incentives created by the gas cap, nor that there
9
Energy Policy 188 (2024) 114092
may be other reasons for this. For example, in a year characterised by
a persistent drought, the low use of hydropower during the summer
months could explain part of the shift to the combined cycle of the
space left in electricity generation, independently of the introduction
of the gas cap.
Fig. 10 offers clues that may support this possible explanation. From
June 15 onward there is a divergent behaviour in the weight that each
technology represents in the total daily energy generated. On the one
hand, hydro and cogeneration show a significant drop, while wind
and solar do not seem to show any particular reaction beyond what
can be attributed to mere seasonality. This divergence between groups
would be a strong candidate to be part of the effects created by the gas
cap’s incentives. Taking for instance the drop in cogeneration, there are
strong reasons to consider that its lower use since the start of the cap
may indeed be due to a direct consequence of the design of the measure
itself: the fact that cogeneration plants have been excluded from the
compensation deriving from the cap may have created incentives to
divert part of the production previously generated by cogeneration
towards combined cycle power plants. Under the measure, part of these
plants would operate in the new regulatory context at a loss, which has
led to the paralysis of a significant portion of the cogeneration fleet.15
As for its effect on gas consumption, although it is true that these
plants also use gas in production, they do so quite efficiently: while
combined cycle plants have an efficiency of 50%–60%, cogeneration
plants tend to have an efficiency of 90%. Thus, it seems at least a priori
could contribute to reducing gas consumption.
However, concerning hydropower, there are reasons to believe that
we may be experiencing a combined effect of several factors operating
in the same direction, making it impossible to differentiate the actual
weight attributable to each. The cap and a concurrent drought that
occurred during the summer of 2022 in Spain could explain part of
the reduction in generation by this technology during the summer
months, conflating both effects. This does not imply that there were
no incentives to reduce hydro generation in favour of combined cycle
power plants. Still, it does mean that an undetermined part of the
increase in the latter, observed in the previous estimate, is not due
to the cap but to the climate event. The combination of these two
factors suggests that hydropower generation would have been different
without them.
Because of these doubts, it is possible that the variables used in
our model to capture the causal effect are not fully exogenous. For
15
As a matter of fact, the Spanish Cogeneration Association (ACOGEN)
called for cogeneration to be included in the compensatory mechanism. By
early September, Prime Minister Pedro Sánchez announced a modification to
the regulation of the cogeneration remuneration system to allow facilities that
so wish to temporarily waive it and receive the adjustment resulting from the
Iberian exception. The Council approved this regulatory change of Ministers on
September 20, so there will likely be an upturn in the participation of these
technologies in electricity generation. This could explain why the effect on
increased usage of combined-cycle power plants appears to fade away from
October 2022 onwards.
M. Hidalgo-Pérez et al. Energy Policy 188 (2024) 114092

Fig. 8. Estimated causal effect for each model with 90% of significant interval. Combined cycles generation.
Note: The blue results refer to the models that include electricity generation from wind, solar plants and hydropower, while the red ones pertain to the use of meteorological
variables and the total system generation. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Source: Own elaboration.

Fig. 9. Estimated causal effect for each model. Estimated increase in electricity generation in combined cycles centrals.
Source: Own elaboration.

Fig. 10. Weight of the different technologies in the total amount generated in 2019, 2021 and 2022.
Note: The 7-day moving average is calculated from March 1 to August 31. The lightest colour stands for 2019, the darkest for 2022. (For interpretation of the references to colour
in this figure legend, the reader is referred to the web version of this article.)
Source: Own elaboration.

this reason, we present the same causal estimation but using only It can be observed that the effect remains positive, although slightly
meteorological variables and total generation as the red lines in Fig. 8. smaller and even insignificant when MIBGAS is used as the reference

10
M. Hidalgo-Pérez et al.
Fig. 11. Use of the Spain–France interconnection.
Source: Own elaboration from REE.es data.
gas price. The results vary between 25% and 10% when using TTF or
the weighted index, and are not significant when using when MIBGAS
is used
Although, as mentioned above, we consider the TTF or the weighted
index to be the best indicators of gas prices, for which the effect is
positive and significant, the non-significant effect in the case of MIBGAS
should not go unnoticed. However, in the case of the TTF, the changes
in our estimate of the causal effect of gas production are higher than
those of the VPSC price. This implies some evidence of the partial
endogeneity of wind and photovoltaic production in the modelling of
CCGT production.
6. Current impact in other countries: possible leakage from the
system
Beyond the impact of the mechanism on prices and the generation
mix in Spain, it was widely speculated among policymaking circles how
its implementation may have a significant impact in France, the country
that holds virtually all European energy interconnections towards the
Iberian peninsula. Although the proposal drafted by Spain and Portugal
initially envisaged a different wholesale price for interconnections, the
European Commission finally ruled out this possibility to avoid restrict-
ing cross-border trade, or discriminate between consumers. Thus, the
lower prices in the Iberian market, together with a series of unsched-
uled stoppages in the French nuclear fleet that took place simultaneous
to the cap implementation, and hydroelectric production at a minimum
due to the aforementioned drought, could have increased the import
demand from France.
Fig. 11 gives us an idea of the magnitude of the change in the
trade balance between Spain and France. Since the entry into effect
of the cap, Spain became a net exporter. In particular, exports more
than doubled in 2022 compared to 2021 and imports almost halved.
This completely reversed the traditional trade balance between the two
countries, while trade itself increased by 56.3%.
This effect could have been even greater had export capacity not
been limited. Export capacity between France and Spain was reduced
by an average of 30% since June 15 compared to the average values
recorded up to that date. Given that the available interconnections
were used at maximum existing capacity since the mechanism became
operational, exports would most likely have been even higher had they
not been restricted.
Assessing the possible impact of the gas cap on exports to France
beyond observational indicative evidence requires adopting an indirect
strategy. First, we estimate a model for the wholesale spot prices (not
VPSC) that would have prevailed in Spain in the absence of the cap
for TTF, MIBGAS and the 80/20 combination. Results can be found in
11
Energy Policy 188 (2024) 114092
Tables A.5 and A.6. Once obtained, we contrast the three counterfactual
wholesale prices in Spain with the French spots. The intuition behind
this exercise is to assume that exports of electricity to France are due to
a lower spot price in our country and, by estimating the counterfactual,
we can visualise whether, even without the existence of the cap, Span-
ish prices would have been cheaper, specifically cheaper than those in
France. We should not forget that the price difference does not allow
us to fully distinguish between the cap effect and the seasonal effects
already mentioned, which dominated the French electricity market
during the months under analysis.
Fig. 12 shows the estimated effect of the gas cap on, in this case, the
spot price for Spain. Once again, it can be seen that the cap led to a
reduction in the spot price in Spain. This reduction is greater than that
of the VPSC because, as we know, the latter includes compensation for
power plants that use gas for electricity generation. With these results,
we can estimate the counterfactual spot price for Spain in the absence
of a gas cap and compare it with the observed one. This is illustrated
in Fig. 13 for each of the gas prices used.
Finally, we compare the three counterfactual spot price series for
Spain with the observed price for France. This comparison is shown in
Fig. 14. It appears that the French price would have been higher than
the Spanish price for most of the weeks since June 15 independent of
the gas price we use.
What is striking, however, is the marked change in the behaviour
of French prices, particularly since the introduction of the gas cap.
Although it could also be due to other reasons that are difficult to
model, we cannot rule out the possibility that the cap may have led
to a strategic shift in the production of the French electricity system.
Let us not forget that the technical shutdown of a number of nuclear
power plants in France came into effect at the end of April, which
would explain why the remaining operating plants were able to more
than cover demand on weekends but were unable to do so adequately
on other days. It is therefore reasonable to consider that the increase
in exports to France was not only and exclusively a direct consequence
of the increase in spot price differentials between the two countries,
since these incentives would most likely have existed even without
the cap, but also due to a French shift in response to the cap. In
any case, it should be borne in mind that the French system is highly
interconnected and therefore does not set prices in isolation.
It is certainly difficult to discern conclusively whether the gas cap
has caused changes in export flows or not. But what is unquestionable
(and mechanic) is that it has generated an economic benefit for French
and Portuguese consumers. And, in any case, the result observed is that
French imports and their price have clearly changed since the measure
came into effect. This would support the fear expressed by Eicke et al.
(2022) of ‘‘leakage’’ to non-member states of the money spent on
M. Hidalgo-Pérez et al. Energy Policy 188 (2024) 114092

Fig. 12. Causal impact of gas cup in wholesale price in Spain.

Note: The blue results refer to the models that include electricity generation from wind and solar plants, while the red ones pertain to the use of meteorological variables and the
total system generation. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Source: Own elaboration.

Fig. 13. Observed wholesale price in Spain and estimated price without the gas cap.
Source: Own elaboration.

Fig. 14. Estimated wholesale price for Spain and observed for France. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version
of this article.)
Source: Own elaboration from REE.es data.

12
M. Hidalgo-Pérez et al.
Fig. 15. Causal impact of gas cap in electricity exports to France.
Source: Own elaboration.
Fig. 16. Causal impact of gas cap in electricity exports to France.
Source: Own elaboration.
compensation. A leakage that may be significantly higher than that
observed between Spain and France simply because the connections
from EU countries to non-EU countries are more intense.
It is with this possibility in mind, and with the aim of understanding
whether there is evidence to support it, that we carry out a final
exercise, this time to analyse the impact on the series of exports from
Spain to France. Tables A.7 and A.8 show the regression used in the
estimation of the causal impact in exports and Fig. 15 the causal impact
using both sets of covariables.
While the use of TTF or weighted gas prices shows evidence of an
increase in exports of between 8% and 35%, with MIBGAS as the ref-
erence price, the evidence is inconclusive when renewable production
is used as a covariate, but not when meteorological variables are used.
However, as observed in Fig. 16, this impact seems to concentrate
only between the months of August and November, coinciding with
the months with lower nuclear and hydroelectric generation in France
during 2022.
13
Energy Policy 188 (2024) 114092
7. Conclusion and policy implications
According to our estimates based on a causal identification model,
the exceptional gas cap for electricity generation in Spain and Portugal,
the most drastic measure taken in Europe to deal with the escalation
in energy prices following the Russian invasion of Ukraine, had a
measurable effect. Prices of the regulated VPSC tariff were, on average,
between 20% and 28% lower than they would have been without the
measure. This lower price, despite the compensation included in the
bill, has resulted in average savings of 129e/MWh per day since June
15 until the end of December 2022.
However, the introduction of the cap may have had other unin-
tended effects. The first of these is a significant increase in the use of
combined cycle power plants at the expense of a reduction in the use of
(non-CO2 emitting) hydro and (more efficient) CHP plants. Although it
is not so evident that, particularly in the case of hydro, the substitution
of these technologies by combined cycle plants is exclusively caused
by the cap, it is no less true that the introduction of the measure in a
M. Hidalgo-Pérez et al. Energy Policy 188 (2024) 114092

particular context that raises incentives to burn gas may be amplifying Appendix
the undesired effects described above. This highlights the inherent
trade-offs policymakers must navigate between short-term economic
relief and long-term decarbonisation. See Tables A.1–A.8.
Finally, while the introduction of the cap may have increased the
incentives to export to France, these would have remained in place even
without the cap. However, we cannot rule out that France’s decisions Table A.1
on its electricity market production may have been equally strategic OLS regressions for VPSC. Results for different gas price models specifications using
given the existence of an implicit subsidy, as the compensation to eolic and photovoltaic production.
combined cycle plants is paid mostly by Spanish consumers. An exercise TTF MIBGAS Weighted
has been conducted, and there is some evidence that the gas cap may const 21.66*** 36.30** 22.64***
have contributed to boosting exports. Therefore, it is feasible that the (1.91) (14.20) (1.88)
Gas price (TTF)𝑡−1 1.21***
increase in exports to France is due both to a lower relative price
(0.09)
created by the cap and to a change in the neighbouring country’s Gas price (TTF)𝑡−2 −0.95***
strategy to take advantage of this eventual situation. (0.09)
Our methodology provides a robust causality analysis through the Gas price (MIBGAS)𝑡−1 1.39***
utilisation of Bayesian structural time series models, a novel approach (0.12)
in this field of energy economics. We have not only quantified the Gas price (MIBGAS)𝑡−2 −0.18
(0.14)
financial impact on consumers but also expanded the analysis to in- Gas price (Weighted)𝑡−1 1.49***
clude the effects on energy sources, effectively capturing both intended (0.09)
and unintended impacts of the policy. These contributions offer a Gas price (Weighted)𝑡−2 −1.17***
rigorous framework for evaluating similar energy market interventions (0.09)
in the future, illustrating how price-based policy mechanisms can have VPSC𝑡−1 0.78*** 0.46*** 0.76***
(0.01) (0.03) (0.01)
far-reaching implications beyond immediate economic effects.
Wind farms −0.07*** −0.20*** −0.07***
It is crucial to acknowledge the limitations of our study. Specif- (0.00) (0.02) (0.00)
ically, the method had difficulty in discerning the direct effects on Solar Photovoltaic −0.06** −0.22*** −0.06***
the increased use of combined cycle plants and on exports to France (0.02) (0.08) (0.02)
due to temporally correlated confounding factors. These limitations Reform 1 10.42*** 9.94***
(2.00) (1.97)
suggest caution in interpreting the study’s results as definitive evidence
Reform 2 5.12** −12.61 3.06
of causality in these particular areas. (2.39) (12.74) (2.35)
Given these findings and limitations, future research should aim R-squared 0.94 0.90 0.95
to better understand the nuanced impacts of such policies on the en- R-squared Adj. 0.94 0.90 0.94
ergy mix and on international trade relations. Exploring heterogeneous No. observations 3214 479 3214
effects on different types of households can also enrich our understand- Note: The dependent variable is VPSC. Each regression uses a gas price indicated in
ing of the policy’s broader socio-economic implications. The quest to the name of each column. ‘Weighted’ refers to a weighted average price, with 80% for
achieve both economic stability and environmental sustainability in the TTF and 20% for MIBGAS. All regressions include controls for days of the week and
months of the year.
energy sector is a complex and urgent challenge, one that demands
Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
further rigorous, empirical investigation.

Table A.2
CRediT authorship contribution statement
OLS regressions for VPSC. Results for different gas price models specifications using
meteorological and total generation variables.
Manuel Hidalgo-Pérez: Conceptualization, Investigation, Method-
TTF MIBGAS Weighted
ology, Software, Supervision, Writing – original draft, Writing – review
const 23.38*** 22.29 25.30***
& editing. Natalia Collado: Conceptualization, Investigation, Super-
(4.89) (23.49) (4.82)
vision, Writing – original draft, Writing – review & editing,. Jorge Gas price (TTF)𝑡−1 1.17***
Galindo: Conceptualization, Supervision. Ramón Mateo: Conceptual- (0.09)
ization, Writing – original draft. Gas price (TTF)𝑡−2 −0.93***
(0.09)
Declaration of competing interest Gas price (MIBGAS)𝑡−1 1.33***
(0.12)
Gas price (MIBGAS)𝑡−2 −0.19
The authors declare the following financial interests/personal rela- (0.14)
tionships which may be considered as potential competing interests: Gas price (Weighted)𝑡−1 1.46***
MANUEL HIDALGO reports financial support was provided by ESADE (0.09)
ECPOL. MANUEL HIDALGO reports a relationship with ESADE ECPOL Gas price (Weighted)𝑡−2 −1.16***
(0.09)
that includes: consulting or advisory. If there are other authors, they
VPSC𝑡−1 0.78*** 0.45*** 0.76***
declare that they have no known competing financial interests or (0.01) (0.04) (0.01)
personal relationships that could have appeared to influence the work average temperature −72.98 −48.55 −58.88
reported in this paper. (85.95) (350.34) (84.51)
precipitation −0.11 −2.20** −0.10
Data availability (0.21) (0.93) (0.21)
minimum temperature 36.53 22.35 29.43
(42.98) (175.12) (42.26)
Data will be made available on request. maximum temperature 35.70 24.66 28.70
(42.98) (175.19) (42.26)
Acknowledgements average wind speed −11.31*** −38.55*** −11.03***
(1.61) (6.94) (1.59)
The authors would like to thank Pedro Linares, Natalia Fabra and maximum wind gust 1.04* 3.31 0.96*
(0.55) (2.40) (0.54)
Isidoro Tapia and two anonymous reviewers for their comments, which
have made a decisive contribution to the analysis presented here. (continued on next page)

14
M. Hidalgo-Pérez et al. Energy Policy 188 (2024) 114092

Table A.2 (continued). Table A.4 (continued).

TTF MIBGAS Weighted TTF MIBGAS Weighted
insolation 0.14 −3.72** 0.11 Gas price (MIBGAS)𝑡−2 −0.19
(0.36) (1.67) (0.35) (0.14)
total demand 0.03*** 0.13*** 0.02*** Gas price (Weighted)𝑡−1 1.46***
(0.01) (0.03) (0.01) (0.09)
Reform 1 8.28*** 7.63*** Gas price (Weighted)𝑡−2 −1.16***
(1.83) (1.80) (0.09)
Reform 2 4.33* −22.41* 2.32 Average temperature −72.98 −48.55 −58.88
(2.38) (12.51) (2.35) (85.95) (350.34) (84.51)
R-squared 0.94 0.91 0.95 Precipitation −0.11 −2.20** −0.10
R-squared Adj. 0.94 0.90 0.94 (0.21) (0.93) (0.21)
No. observations 3214 479 3214 Minimum temperature 36.53 22.35 29.43
(42.98) (175.12) (42.26)
Note: The dependent variable is VPSC. Each regression uses a gas price indicated in
Maximum temperature 35.70 24.66 28.70
the name of each column. ‘Weighted’ refers to a weighted average price, with 80% for
(42.98) (175.19) (42.26)
TTF and 20% for MIBGAS. All regressions include controls for days of the week and
Average wind speed −11.31*** −38.55*** −11.03***
months of the year.
(1.61) (6.94) (1.59)
Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
Maximum wind gust 1.04* 3.31 0.96*
(0.55) (2.40) (0.54)
Table A.3 Insolation 0.14 −3.72** 0.11
OLS regressions for combined cycles generation. Results for different gas prices (0.36) (1.67) (0.35)
specifications. Total demand 0.03*** 0.13*** 0.02***
TTF MIBGAS Weighted (0.01) (0.03) (0.01)
Reform 1 8.28*** 7.63***
const 19.69*** 19.60*** (1.83) (1.80)
(3.21) (3.21) Reform 2 4.33* −22.41* 2.32
Combined cycles𝑡−1 0.75*** 0.59*** 0.75*** (2.38) (12.51) (2.35)
(0.01) (0.03) (0.01) R-squared 0.94 0.91 0.95
Gas price (TTF)𝑡−1 −0.03 R-squared Adj. 0.94 0.90 0.94
(0.14) No. observations 3214 479 3214
Gas price (TTF)𝑡−2 0.12
(0.14) Note: The dependent variable is combined cycles generation. Each regression uses a gas
Gas price (MIBGAS)𝑡−1 0.19 price indicated in the name of each column. ‘Weighted’ refers to a weighted average
(0.16) price, with 80% for TTF and 20% for MIBGAS. All regressions include controls for days
Gas price (MIBGAS)𝑡−2 0.00 of the week and months of the year.
(0.16) Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
Gas price (Weighted)𝑡−1 0.08
(0.15) Table A.5
Gas price (Weighted)𝑡−2 0.01 OLS regressions for wholesale (spot) price. Results for different gas prices specifications.
(0.15)
TTF MIBGAS WEIGHTED
Wind farms −0.17*** −0.39*** −0.17***
(0.01) (0.03) (0.01) const 8.67*** 29.92** 8.53***
Solar Photovoltaic 0.15*** −0.33*** 0.16*** (1.12) (11.91) (1.11)
(0.04) (0.11) (0.04) Spot price𝑡−1 0.87*** 0.78*** 0.87***
Hydropower −0.02 −0.01 −0.02 (0.01) (0.03) (0.01)
(0.02) (0.15) (0.02) Gas price (TTF)𝑡−1 0.87***
Reform 1 −2.50 84.04*** −2.69 (0.06)
(3.14) (25.63) (3.15) Gas price (TTF)𝑡−2 −0.85***
Reform 2 12.04*** 9.15 12.39*** (0.06)
(3.84) (16.88) (3.87) Gas price (MIBGAS)𝑡−1 1.10***
R-squared 0.84 0.84 0.84 (0.10)
R-squared Adj. 0.84 0.83 0.84 Gas price (MIBGAS)𝑡−2 −0.85***
No. observations 3214 479 3214 (0.10)
Gas price (Weighted)𝑡−1 1.07***
Note: The dependent variable is production in combined cycles centrals. Each regression
(0.06)
uses a gas price indicated in the name of each column. ‘Weighted’ refers to a weighted
Gas price (Weighted)𝑡−2 −1.05***
average price, with 80% for TTF and 20% for MIBGAS. All regressions include controls
(0.06)
for days of the week and months of the year.
Wind farms −0.05*** −0.12*** −0.05***
Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
(0.00) (0.01) (0.00)
Solar Photovoltaic −0.04** −0.14** −0.04**
(0.01) (0.07) (0.01)
Reform 1 9.51*** 9.19***
(1.42) (1.40)
Table A.4
Reform 2 8.61*** −0.72 7.73***
OLS regressions for combined cycles generation. Results for different gas price models
(1.69) (10.59) (1.66)
specifications and using meteorological and total generation variables.
R-squared 0.95 0.87 0.96
TTF MIBGAS Weighted R-squared Adj. 0.95 0.86 0.96
const 23.38*** 22.29 25.30*** No. observations 3214 479 3214
(4.89) (23.49) (4.82)
Note: The dependent variable is the Spanish wholesale price (spot price). Each
Gas price (TTF)𝑡−1 1.17***
regression uses a gas price indicated in the name of each column. ‘Weighted’ refers
(0.09)
to a weighted average price, with 80% for TTF and 20% for MIBGAS. All regressions
Gas price (TTF)𝑡−2 −0.93***
include controls for days of the week and months of the year.
(0.09)
Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
Gas price (MIBGAS)𝑡−1 1.33***
(0.12)

15
M. Hidalgo-Pérez et al. Energy Policy 188 (2024) 114092

Table A.6 Table A.7 (continued).

OLS regressions for wholesale (spot) price. Results for different gas price models TTF MIBGAS WEIGHTED
specifications and using meteorological and total generation variables.
Wind farms 54.77*** 34.77*** 54.78***
TTF MIBGAS WEIGHTED
(2.52) (7.47) (2.52)
const 19.14*** 93.15*** 19.77*** Solar Photovoltaic 68.17*** 46.97 68.46***
(3.27) (19.60) (3.22) (12.39) (35.67) (12.37)
Spot price𝑡−1 0.88*** 0.77*** 0.88*** Reform 1 −4810.61*** −4702.93***
(0.01) (0.03) (0.01) (1100.26) (1103.79)
Gas price (TTF)𝑡−1 0.84*** Reform 2 494.18 2494.60 894.46
(0.06) (1316.51) (5459.68) (1323.82)
Gas price (TTF)𝑡−2 −0.83*** R-squared 0.77 0.73 0.77
(0.06) R-squared Adj. 0.77 0.71 0.77
Gas price (MIBGAS)𝑡−1 1.07*** No. observations 3214 479 3214
(0.10)
Note: The dependent variable is the exports to France. Each regression uses a gas price
Gas price (MIBGAS)𝑡−2 −0.84***
indicated in the name of each column. ‘Weighted’ refers to a weighted average price,
(0.10)
with 80% for TTF and 20% for MIBGAS. All regressions include controls for days of
Gas price (Weighted)𝑡−1 1.05***
the week and months of the year.
(0.06)
Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
Gas price (Weighted)𝑡−2 −1.03***
(0.06)
Average temperature −39.72 −81.91 −32.12
(57.63) (288.10) (56.87) Table A.8
Precipitation −0.05 −0.70 −0.05 OLS regressions for exports to France. Results for different gas prices specifications.
(0.14) (0.77) (0.14)
TTF MIBGAS WEIGHTED
Minimum temperature 20.24 40.68 16.45
(28.82) (144.02) (28.44) Const −39245.00*** −61791.56*** −39231.29***
Maximum temperature 18.81 38.94 15.02 (2690.05) (10288.81) (2690.50)
(28.82) (144.07) (28.44) Exports𝑡−1 0.66*** 0.58*** 0.66***
Average wind speed −8.86*** −21.73*** −8.74*** (0.01) (0.03) (0.01)
(1.08) (5.71) (1.07) Spot price France 39.15*** 17.90 43.09***
Maximum wind gust 1.02*** 1.91 1.01*** (8.97) (14.16) (9.13)
(0.37) (1.97) (0.36) Spot price France𝑡−1 −20.49** 12.45 −17.99*
Insolation 0.64*** 0.22 0.66*** (9.05) (14.08) (9.22)
(0.24) (1.37) (0.23) Gas price (TTF)𝑡−1 −57.48
Total demand 0.01* −0.01 0.01 (50.24)
(0.00) (0.02) (0.00) Gas price (TTF)𝑡−1 69.87
Reform 1 7.69*** 7.31*** (49.74)
(1.28) (1.26) Gas price (MIBGAS𝑡−1 −47.22
Reform 2 7.99*** −8.01 7.12*** (52.89)
(1.68) (10.34) (1.65) Gas price (MIBGAS)𝑡−1 1.95
R-squared 0.96 0.87 0.96 (52.26)
R-squared Adj. 0.95 0.87 0.96 Gas price (Weighted)𝑡−1 −82.41
No. observations 3214 479 3214 (53.28)
Gas price (Weighted)𝑡−1 75.30
Note: The dependent variable is the Spanish wholesale price (spot price). Each
(52.80)
regression uses a gas price indicated in the name of each column. ‘Weighted’ refers
Average temperature 48 051.58 167 968.60 46 723.74
to a weighted average price, with 80% for TTF and 20% for MIBGAS. All regressions
(46896.01) (142873.38) (46898.29)
include controls for days of the week and months of the year.
Precipitation 250.03** 449.25 237.72**
Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
(116.12) (381.67) (116.14)
Minimum temperature −23988.69 −82910.46 −23315.05
(23446.84) (71415.52) (23447.95)
Maximum temperature −23870.62 −83906.12 −23203.74
Table A.7 (23451.66) (71444.45) (23452.79)
OLS regressions for exports to France. Results for different gas prices specifications. Average wind speed 5673.16*** 3406.55 5610.52***
TTF MIBGAS WEIGHTED (880.89) (2839.03) (881.13)
Maximum wind gust −562.21* −489.33 −530.19*
const −6753.67*** −3596.64 −6708.59***
(297.88) (977.70) (298.01)
(837.86) (5991.36) (838.96)
Insolation −59.79 248.92 −61.00
Exports𝑡−1 0.66*** 0.66*** 0.66***
(193.38) (679.83) (193.38)
(0.01) (0.03) (0.01)
Total demand 43.70*** 74.29*** 43.43***
Spot price France 50.80*** 28.53* 53.78***
(3.79) (13.33) (3.79)
(9.06) (14.63) (9.22)
Reform 1 531.08 663.78
Spot price France𝑡−1 −15.51* 18.32 −13.43
(978.78) (982.63)
(9.16) (14.58) (9.33)
Reform 2 752.76 8105.49 1232.88
Gas price (TTF)𝑡−1 77.18
(1293.28) (5205.71) (1301.40)
(50.49)
R-squared 0.78 0.76 0.78
Gas price (TTF)𝑡−1 −74.24
R-squared Adj. 0.78 0.74 0.78
(51.10)
No. observations 3214 479 3214
Gas price (MIBGAS)𝑡−1 −60.84
(54.53) Note: The dependent variable is the exports to France. Each regression uses a gas price
Gas price (MIBGAS)𝑡−1 −3.97 indicated in the name of each column. ‘Weighted’ refers to a weighted average price,
(54.23) with 80% for TTF and 20% for MIBGAS. All regressions include controls for days of
Gas price (Weighted)𝑡−1 −89.59* the week and months of the year.
(54.16) Standard errors in parentheses *p<0.1, **p<0.05, ***p<0.01.
Gas price (Weighted)𝑡−1 76.51
(53.58)

16
M. Hidalgo-Pérez et al.
References
Abadie, A., Diamond, A., Hainmueller, J., 2010. Synthetic control methods for compar-
ative case studies: Estimating the effect of California’s tobacco control program. J.
Amer. Statist. Assoc. 105, 493–505. http://dx.doi.org/10.1198/jasa.2009.ap08746.
Arroqui, M., 2022. La UE avisa a españa del aumento del consumo por su tope al
gas. https://www.larazon.es/economia/20221001/nxzylv7265hkxpgbq56wvm4ufe.
html.
Brito, P., 2022. La excepción ibérica a debate una oportunidad perdida? URL https:
//ungradoymedio.blogspot.com/2022/06/la-excepcion-iberica-debate-una.html.
Brodersen, K.H., Gallusser, F., Koehler, J., Remy, N., Scott, S.L., 2015. Inferring causal
impact using bayesian structural time-series models. Ann. Appl. Stat. 9, 247–274.
http://dx.doi.org/10.1214/14-AOAS788.
CNMC, 2022. Boletín de indicadores eléctricos de junio de 2022. URL https://www.
cnmc.es/sites/default/files/4195294.pdf.
CNMC, 2023. Boletín de indicadores eléctricos. URL https://www.cnmc.es/sites/
default/files/4623269.pdf.
Eicke, A., Hirth, L., Maurer, C., Mühlenpfordt, J., Schlecht, I., 2022. The iberian
electricity market intervention does not work for europe. VoxEU org.
European Commission, 2022. Communication from the commission to the European
parliament, the European council, the council, the European economic and social
committee and the committee of the regions security of supply and affordable en-
ergy prices: Options for immediate measures and preparing for next winter. https://
eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX, URL https://eur-lex.europa.
eu/legal-content/EN/TXT/?uri=CELEX%3A52022DC0138&qid=1649253932345.
17
Energy Policy 188 (2024) 114092
Hernán, M.A., Robins, J.M., 2023. Causal Inference: What If, July, 2023 Chapman &
Hall/CRC, Boca Ratón.
Hidalgo-Pérez, M.A., Van-Baumberguen, N.C., Jorge, G., Mateo, R., 2022. Los
efectos del tope al gas en los precios, la inflación y el consumo seis meses
después. Madrid, pp. 1–15. http://dx.doi.org/10.56269/MHP20230131, URL
file:///C:/Users/Usuario/Downloads/AAFF_ESP_EsadeEcPol_Insight42_Los-Efectos-
del-tope-del-gas_corregido.pdf.
Independiente, E., 2022. España deberá vender la electricidad a francia al precio
topado por la ’excepción ibérica’. https://www.elindependiente.com/economia/
2022/05/26/espana-debera-vender-la-electricidad-a-francia-al-precio-topado-por-
la-excepcion-iberica/.
Mosquera-López, S., Uribe, J.M., Manotas-Duque, D.F., 2017. Nonlinear empirical
pricing in electricity markets using fundamental weather factors. Energy 139,
594–605.
Robinson, D., Arcos-Vargas, A., Tennican, M., nez, F.N., 2023. The iberian exception:
An overview of its effects over its first 100 days. arXiv preprint 2309.02608.
Salas, R., Jerez, M., Álvarez, F., 2022. Actualización indicador ICAE del impacto
del tope del gas en el precio de la electricidad. ICAE Blog URL https://blog-
icae.blogspot.com/2022/09/actualizacion-indicador-icae-del_30.html.
Sancha, J.L., 2022. Balance económico de 200 días de aplicación del tope al precio del
gas. Cuadernos Energía 71, 46–57.
Sgaravatti, G., Tagliapietra, S., Zachmann, G., 2021. National policies to shield
consumers from rising energy prices. Bruegel Datasets 4.